This is my contribution to the #ClimateStrike. For the past 8 years or so, I’ve volunteered as energy coordinator of the Informatics Forum. My mission has been to reduce energy consumption in the Forum, both by encouraging occupants to change their behaviour (e.g. by turning off lights and equipment) and to identify changes in building operation. It’s been a steep learning curve, as my academic expertise is computational neuroscience, not engineering.
Here are some reflections on what I’ve learned:
- Modern buildings are complicated, with a big chunk of energy being consumed behind the scenes on services such as ventilation.
- The details of construction and operation really matter to efficient operation. I have had the privilege of working with Bill Bordass, helping with a “Post Occupancy Evaluation” of the building. He explains what’s wrong with how we build now and how we can do better.
- Controls and control engineers are crucial. The University’s control engineers were able to save about 15% of our electricity consumption by automating the opening and closing of a vent and adjusting the controls. Control engineers are my unsung heroes.
- The spec of the building is crucial. Somehow we ended up having a very oversized uninterruptible power supply (UPS) that gobbled another 15% of our power, but it was never clear who wanted it in the first place! (Thankfully we’ve now replaced it and another UPS – the signs are that we’re saving a decent amount of electricity.)
- Data can be quite helpful. We have automated readings of overall electricity consumption and how much is used by servers and offices. Though one of the biggest insights into energy consumption came from 2 manual readings of a meter in the plant room!
- Academic buildings are an energy nightmare compared to standard office buildings. We stay in them at all hours, so the automated controls on lights in corridors don’t get much of a chance to turn the lights off.
Here’s what we (and the University) could think about:
- Is it crucial that all our servers stay on during a power cut? When it next comes to replacing the UPS, we could perhaps get a smaller one if demand for it is lower. We should really be checkpointing our jobs anyway.
- Server energy consumption use has been increasing (perhaps something to do with deep learning and GPUs?) Can we make our code any more energy efficient? Can we teach energy efficient coding?
- Can we replace the fluorescent lights with LEDs, and perhaps tweak the controls?
- Can we reduce the need for travel and travel more often by train? Of course, plane is often cheaper than plane, even for UK mainland travel, but train fares, even first class, can be reasonable in advance and the University expenses policy says that “a non-standard class rail fare is allowable where the claimant plans to work for the duration of the journey“.
- The ventilation in some offices is still not good, and could, I think (I’m not a building engineer), be modified without increasing energy consumption on ventilation – and might perhaps allow us to reduce consumption.
I’m sure there’s more to think about – comments welcome.
“Can we make our code any more energy efficient? Can we teach energy efficient coding?”
For a few years now I’ve been saying that algorithms people can increase the perceived “impact” of their work by couching it as contributing to energy efficiency, particularly for algorithms commonly used in datacentres. And for those who understand macroeconomics and the second law of thermodynamics, energy efficiency is another term for energy productivity, so it’s also a contribution to economic productivity.
“Can we make our code any more energy efficient?”
Even though this seems to point to the right direction, maybe it doesn’t: when people found out ways to burn coal more efficiently, they started polluting more – https://en.wikipedia.org/wiki/Jevons_paradox
Also, let’s keep the big picture in mind: switching off the light may make us feel better and save a marginal amount of energy, but saving the planet includes putting up against a tradition of expending enormous amounts of energy for building new stuff, transporting cargo & humans ever more and finding new ways to pollute. Remember that the energy consumption trend is still increasing globally (!) — here are some nice charts: https://ourworldindata.org/energy-production-and-changing-energy-sources
Luckily, in UK & EU consumption is declining on aggregate, but probably not fast enough. And unfortunately we have got used to travelling and importing a lot, so specifically consumption for the transport sector is increasing in the UK — https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/820843/Energy_Consumption_in_the_UK__ECUK__MASTER_COPY.pdf.
Disclaimer: I switch off the light (which makes me feel better) and I have put up zero practical fight against the erection of the Bayes Centre and the (opinion ahead) outright useless shopping mall on Leith Street. I also fly much more often than what should be the ethical standard by now, i.e. one flight per year.
Some good points.
Let’s not forget that the University also adds substantial CO2 to the atmosphere from the flights we all take as part of our job. We need action to encourage and support faculty in using land-based travel, and also help to plan business travel sensibly, to avoid hops across the atlantic for a couple of days. We have the Tech to support remote participation in conferences, we should be using it more.
Yet another comment on “Can we make our code any more energy efficient?”
A recent preprint on arxiv advocates for using energy efficiency as an evaluation criterion along with accuracy measures in training neural networks. However, it remains to be seen whether it will be adopted by the community.
First off, I’m all for saving energy where we can, but would it matter how much energy we used, if it all came from renewable/green sources?
A couple of comments:
1. Can you clarify exactly how much energy was saved by the recent removal of the 500kVA ‘building’ UPS?
2. You write: ‘Is it crucial that all our servers stay on during a power cut? When it next comes to replacing the UPS, we could perhaps get a smaller one if demand for it is lower.’
In fact the function of the server room UPS isn’t really to ‘keep all servers on during a power cut.’ Rather, it’s to provide conditioned power to servers (i.e. to guard the hardware against spikes, surges, sags, and brownouts), and to allow an orderly shutdown in the event of a prolonged power cut.
Generally, in the UK, 95% of all mains outages last for less than five minutes, with anything longer likely to last for several hours. The new server room UPS has a run-time (when fully-loaded) of 30 minutes. If the objective is really to ‘keep all servers on during a power cut’, we would need to install a back-up generator.
However a good question – and I guess what you were alluding to – is whether some/all servers can be completely removed from the UPS supply. That’s indeed something worth considering.
3. You write: ‘Server energy consumption use has been increasing (perhaps something to do with deep learning and GPUs?) Can we make our code any more energy efficient?’
Certainly, server energy consumption has shot up over the past couple of years mainly on account of the GPU bandwagon. How and why we’re using those things so enthusiastically is a good question. A related question is: how many do we really need?